Low‐level LIBS and Raman data fusion in the context of in situ Mars exploration

• Published in: Journal of Raman spectroscopy 2020-09, Vol.51 (9), p.1682-1701
• Main Authors: Rammelkamp, Kristin, Schröder, Susanne, Kubitza, Simon, Vogt, David S., Frohmann, Sven, Hansen, Peder B., Böttger, Ute, Hanke, Franziska, Hübers, Heinz‐Wilhelm
• Format: Article
• Language: English
• Subjects: Atmospheric conditions; Atmospheric models; Basalt; Carbonates; Computer simulation; Data analysis; data fusion; Data integration; Discriminant analysis; LIBS; Mars; Mars atmosphere; Mars exploration; Multivariate analysis; Perchlorates; Principal components analysis; Raman; Raman spectra; Raman spectroscopy; Salts; Spectroscopy; Spectrum analysis; Sulfates
• ISSN: 0377-0486; 1097-4555
• DOI: 10.1002/jrs.5615

Laser‐induced breakdown spectroscopy (LIBS) and Raman spectroscopy are powerful key techniques for the geoanalytical exploration of extraterrestrial bodies, especially when combined. Their data are complementary, which motivates the question of how it can be best combined to maximize the scientific output. For this study, LIBS and Raman data from pure sulfates and their mixtures as well as from other Mars‐relevant salts such as carbonates, chlorides, perchlorates, and sulfates in a basaltic matrix were measured and investigated. All measurements were performed with miniaturized setups, and LIBS experiments were done in simulated Martian atmospheric conditions. Multivariate data analysis (MVA) techniques such as principal component analysis (PCA) and partial least squares discriminant analysis (PLS‐DA) were employed to evaluate the potential for identifying the sulfates or the salts in the basalt with LIBS and Raman data alone and with their low‐level fused data. We found that low‐level data fusion, that is, combination of LIBS and Raman spectra at the data level, can improve the identification of sulfates and salts. Although the approach of low‐level data fusion aims to use all relevant information from both techniques, we observed that not all benefits from the single models are completely represented by the fused model. The computation and performance of appropriate MVA models are affected by the weighting of the single spectra in the combined one, by the dimensionality of the MVA models, and in case of PLS‐DA, by the given input data. From this study, we conclude that generally, data fusion of LIBS and Raman is an advantage for the identification of unknown samples but that more levels, especially, high‐level data fusion (decision level), should be further investigated. With low‐level LIBS and Raman data fusion, a more comprehensive sample identification is possible. As the complexity of the data is increased, a careful analysis including balancing between the data of both techniques is necessary. Depending on the science goal, the particular approach should be chosen.